Aspect category detection is an interesting task that can help computers understand semantics hidden in the reviews and usually play a key role in many business intelligence applications. The previous researches have demonstrated that the deep learning based methods outperform the lexicon-based methods on the task. Specifically, when applying the attention mechanism to the deep neural network, the performance of the model hassignificantly improved. However, certain reviews can be longer than usual, and thereby encoding these sentences will lead to a long-term memorization burden in LSTM model. To meet the challenge, we proposed a joint model that combines word-level and sentence-level self-attention mechanisms. Consequently, the joint model can deal with both long and short reviews well. We conducted experiments on both SemEval-2014 and SemEval-2016 dataset. The experimental results show the joint model outperforms the state-of-the-art methods
Wang, Siyu; Qiu, Jiangtao; and Hong, Chuanyang, "A Joint Self-Attention Model for Aspect Category Detection in E-Commerce Reviews" (2021). PACIS 2021 Proceedings. 194.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.